auto.NEGOTIA

With the new advancements in Generative AI -- We are automating the procurement process, from getting the requirements from customer to negotiation with the vendors, end-to-end.

Comment

Our project was inspired by a simple question: Could Generative AI redefine the end-to-end procurement process, introducing efficiency and logic into negotiations, generally driven by human intuition? We started with identifying customer requirements, leveraging Gen AI to interpret nuanced requests and translating them into objective criteria, then leading into an logic-based negotiation criteria.

The system is an end-to-end procurement automation platform utilizing a chatbot interface and automated email communication. Here's a breakdown of its functionality: -Customer Interaction: The chatbot engages with customers to gather specific details about the product they require. The chatbot compiles this information into a structured JSON format, capturing the correct specifications of the customer's needs. -Vendor Matching: The system has a predefined list of vendors, each with descriptions of the products they offer. Using the embeddings, the system matches the customer's requirements to suitable vendors from the list using similarity metrics. -Request for Quotation (RFQ): The system automatically generates and sends emails to the matched vendors, requesting quotes, timelines, and additional details. Vendors are prompted to provide information about what they can and cannot deliver, along with any alternative solutions they might offer. -Quote Comparison: Upon receiving responses, the system aggregates and compares the quotes and details like timeline and alternatives from different vendors. It identifies the vendor with the least cost and the shortest delivery timeline as the initial "best vendor" using a weighted sum of the cost, timeline and fulfilled requirements. If the best vendor does not have the least cost and least days to deliver the product across the selection, the negotiation process is initiated. -Negotiation Process: If the initial best vendor does not meet predefined criteria (e.g., price or delivery time), the system initiates a negotiation process. Automated emails are sent to the vendor, requesting adjustments to the price and timeline to meet the desired criteria. The negotiation continues iteratively until the system is satisfied with the vendor's offer. -Final Vendor Selection: Once a vendor meets the criteria or negotiates down to the desired terms, they are selected as the final "best vendor." -Customer Presentation: The system presents the selected vendor's offer to the customer, showcasing the best combination of cost and delivery time. If the customer approves, the procurement process concludes with the customer obtaining the desired product from the best vendor.

We used ChatGPT functions and AI to create a chatbot-driven procurement system. ADA002 embeddings helped match product details to vendor offerings, while Streamlit made the interface user-friendly. The system, built in Python, streamlines communication via automated emails.

Challenges we ran into : The unpredicted behaviour of OpenAI and the identification of the proper system for ranking of vendors

What are we proud of: To pull off all the three cases that we had thought up in the start of the Hackathon was a tough ask, but we managed to do it. We finetuned our chatbots to pick up the different ways in which people place the orders and managed to find a good logic to balance the quotes provided by the vendors. Additionally, our user interface is quite intuitive and features a communication with the user and the different vendors.

What we learned: It is not easy to build an end-to-end product in a few hours! But the main takeaway from this for us was the process of procurement, how to automate it, and the challenges in negotiating using coded logic.

What's next for auto.NEGOTIA: The ides of making negotiation in procurement is quite challenging, so there are always ways to improve the algorithm to adapt to what the customer might value more. We look forward to improving the system by adding more nuances, and getting vendor quotes not only through emails, but also through quote pages that they provide.

Built With

Share this project:

Updates